Guo, H, Wang, X, Cheng, H and Huang, M
A location aided controlled spraying routing algorithm for Delay Tolerant
Networks
http://researchonline.ljmu.ac.uk/id/eprint/7253/
Article
LJMU has developed LJMU Research Online for users to access the research output of the
University more effectively. Copyright © and Moral Rights for the papers on this site are retained by the individual authors and/or other copyright owners. Users may download and/or print one copy of any article(s) in LJMU Research Online to facilitate their private study or for non-commercial research. You may not engage in further distribution of the material or use it for any profit-making activities or any commercial gain.
The version presented here may differ from the published version or from the version of the record. Please see the repository URL above for details on accessing the published version and note that access may require a subscription.
For more information please contact [email protected]
http://researchonline.ljmu.ac.uk/
Citation
(please note it is advisable to refer to the publisher’s version if you
intend to cite from this work)
Guo, H, Wang, X, Cheng, H and Huang, M (2017) A location aided controlled
spraying routing algorithm for Delay Tolerant Networks. Ad Hoc Networks,
66. pp. 16-25. ISSN 1570-8705
A location aided controlled spraying routing
algorithm for Delay Tolerant Networks
Hang Guoa, Xingwei Wangb, Hui Chengc, Min Huanga
aCollege of Computer Science and Engineering, Northeastern University, Shenyang, China {[email protected]; [email protected]}
bCollege of Software, Northeastern University, Shenyang, China [email protected]
cDepartment of Computer Science, Liverpool John Moores University, Liverpool, UK [email protected]
Abstract—Delay Tolerant Networks (DTNs) often suffer from intermittent disruption and variable long delay due to factors such as mobility and energy. In this paper, a Location Aided Controlled Spraying (LACS) routing algorithm is proposed to deal with the challenging issues in DTN routing. Only the routing information carried by the contacted nodes is needed in this algorithm, and there is no need for global networks knowledge and hardware support. The routing process is divided into two stages, i.e., controlled spraying routing stage and single-copy routing stage. The maximum transfer throughput of the contact is checked before each message is forwarded. During the controlled spraying stage, the current node adjusts spraying strategy according to the encounter angle of the contact nodes. During the single-copy stage, a location prediction model based on the semi-Markov process (SMP) is introduced, and the node’s behaviors can be captured both in the temporal and spatial domains with this model. The current node predicts the destination node’s location, and then decides whether to forward the message to target node based on the time used for meeting the destination node. Simulation results show that the proposed algorithm can achieve better performance than the traditional routing schemes of DTNs in terms of delivery ratio, network overhead and transmission delay under both random node movement model and realistic trace scenario.
Keywords–delay tolerant networks; semi-Markov process; location prediction; nodes contact; spraying routing
1.
INTRODUCTION
Delay Tolerant Networks (DTNs) [1] are designed to cope with the challenging conditions in the restricted networks with sparse density, intermittent disruption and limited energy. DTNs have features such as multiple hops, self-organization, and no central administration. In DTNs, an end-to-end path between a source node and a destination node does not exist most of the time, and the messages are opportunistically routed. Therefore, routing is an extremely challenging problem due to the characteristics of DTNs and the traditional network routing techniques cannot work effectively in DTNs [2].
There have been different kinds of routing schemes proposed for DTNs in recent years, and typical schemes include epidemic routing and probabilistic routing. The epidemic routing can achieve high delivery ratio and small delay at the cost of
network bandwidth and buffer space by flooding messages to all the nodes encountered. To reduce the consumption of network resource, it uses different ways to limit the number of message copies forwarded, but it is hard to distinguish between nodes with regard to their probabilities of delivering the messages. The probabilistic routing tries to reduce message overhead and buffer contention by forwarding messages to nodes only with high delivery probabilities. Its difficulty lies in the determination of reasonable delivery probability and the applicability in different scenarios.
In this paper, a novel routing algorithm called Location-Aided Controlled Spraying (LACS) is proposed. Despite the fact that a number of existing protocols have used location-aided methods, there are few routing schemes that consider rational assumptions for DTNs and have low complexity [3] [4]. The proposed scheme combines the advantages of both spraying routing and probabilistic routing. The scheme consists of two stages according to the amount of the message copies: the controlled spraying stage and the single-copy location aided routing stage. If message copies are more than one, the message is forwarded by the spraying method, and the spraying process is controlled by node encounter angle instead of binary mode. By this, the copies can be spread to the whole region more quickly and avoid the slow start problem [5]. When only one copy is left, the LACS will trigger the single-copy routing stage aided by the location information. The location prediction model is based on the SMP, and all the needed information can be obtained from the two contact nodes. The last copy of the message will be forwarded to the node which can reach the destination within the shortest time.
The major contributions of this paper are as follows. (1) A method to estimate the throughput of node contact is proposed. It can help avoid transmitting such messages that could not be finished within the contact time. (2) An angle related adaptable spraying strategy is developed to increase the message propagation efficiency. (3) A location prediction model based on the SMP is devised to find the superior relay node in the one-copy routing stage. (4) Simulations are implemented in both synthetic movement models and realistic traces to demonstrate the superior performance of the proposed LACS scheme over several representative routing algorithms.
The remaining sections of this paper are organized as follows. In Section 2, the state-of-the-art routing algorithms in DTNs are briefly reviewed. In Section 3, the location-aided spraying routing algorithm is proposed and analyzed. The simulation results under different nodes mobility models are presented in Section 4. Finally, Section 5 concludes the paper and points out the future work.
2.
RELATED
WORK
There are many kinds of DTN routing schemes and different ways to categorize them [6]. In this paper, they are categorized into two types: coding-based routing and copy-based routing.
2.1. coding-based routing protocols
Motivated by transmitting large size data, for example, big files in DTNs, coding-based routing protocols are proposed against the limited transmission opportunities by relaying buffers in realistic network settings. In the coding-based routing protocols, the source node converts one generated message into multiple coding blocks and transmits them in the DTNs, and the destination node restores the original message after receiving the certain amount of coding blocks. There are mainly two kinds of coding-based routing protocols: erasure coding and network coding.
In [7], an erasure coding based forwarding algorithm is proposed, which spreads the forwarding responsibility over many nodes while maintaining a fixed network overhead. However, this scheme increases the transmission delay in most cases. In [8], the adaptive erasure coding routing schemes are proposed for interplanetary networks, which are effective even in case of imperfect transmission channel knowledge. The incremental redundancy adaptation (IRA) and partially observable Markov decision process (POMDP) approaches are analyzed in the coding round, and the appropriate decision about coding strategy selection can be taken based on incomplete knowledge. In [9], an erasure coding framework integrated within the DTN architecture is introduced, which extends the bundle protocol specification. In this scheme, the transmission robustness against link interruptions and disruption is improved by the joint use of a packet level coding approach with the custody transfer option. In [10], a routing algorithm is proposed by combining the erasure coding and replication to handle the path failures in DTNs. Both Bernoulli path delivery model and Gaussian path delivery model are discussed to solve the underlying optimization problem.The erasure coding based scheme is robust to failures of a few relay nodes, but it also results in underperformance because of the limited storage capacity and long transmission delay in DTNs.
The network coding is defined as allowing intermediate nodes to not only forward but also combine their incoming independent information flows. The network coding based probabilistic routing is proposed in [11], which can reduce the overhead of probabilistic routing algorithms. The nodes do not simply forward packets which they overheard, but may send out information that is coded over the contents of several packets they received. The network coding based epidemic routing (NCER) scheme is proposed in [12], which transmits a batch of data packets with network coding. In [13], an efficient network coding based protocol (E-NCP) is proposed to optimize packet transmission efficiency, achieving similar transmission delay but with much fewer transmissions. In this scheme, the source node transmits slightly more coded packets such that these coded packets are sufficient to decode the original packets with high probability. The coded packets are referred to as pseudo source packets and disseminated by binary spraying. In [14], an efficient context-aware network coding (CANCO) scheme for DTNs is proposed, in which the nodes move according to non-homogeneous mobility models. The friendliness metric is used to measures how popular a node is while the delivery predictability estimates the probability of reaching another node. The most compelling benefits of network coding may be its robustness and adaptability, which can significantly increase throughput for different traffic patterns. However, network coding also increases communication complexity and is not suitable
for the resource constrained DTNs nodes. Therefore, in our scheme, the coding-based method is not used because we are trying to find a simple but effective method which has low demand for node performance and network resource.
2.2. copy-based routing protocols
In the copy-based routing, the source node generates one message and then injects one or more copies into the networks. The relay nodes will forward this message to the destination node by different routing algorithms. The copy-based routing protocols can be classified into single-copy routing and multi-copy routing, and the latter type is in the majority.
In the single-copy routing schemes, a prediction assisted algorithm is proposed for underwater DTNs [15], which uses the aggressive chronological projected graph (ACPG) to capture the network mobility properties and the common characteristics of near optimal routes. Then based on the guidance from ACPG, the online heuristic protocol is devised by choosing appropriate historical information and forwarding criteria. In [16], several basic single-copy routing protocols are discussed such as direct transmission, randomized routing, utility-based routing with 1-hop diffusion, utility-based routing with transitivity, seek and focus routing, oracle-based routingand so on. In [17], a predict and relay routing algorithm is presented in which the nodes determine the probability distribution of future contact times and choose a proper next hop to improve the end-to-end delivery probability. It assumes that the nodes move around a set of well-visited landmark points and their mobility behaviors are described by time-homogeneous semi-Markov process model. The single-copy routing scheme has excellent performance at the expenditure of network resource, but performs poor at other aspects such as message delivery ratio, delay and so on. This kind of routing scheme is appropriate for the scenario where nodes and networks are extremely resource constrained, so it is not applied in our proposed scheme.
The multi-copy routing is the most discussed case in recent years. The epidemic routing is a typical one, in which nodes exchange all different messages they carry and try their best to transmit messages to destination nodes. However, this will lead to too much network resource expenditure. There are some kinds of methods to avoid message flooding: probabilistic routing, social based routing and controlled epidemic routing.
The probabilistic routing determines whether a message between two contacted nodes is transmitted by a certain algorithm. The Prophet routing algorithm is proposed in [18], which uses history information of previous encounters to estimate delivery probability. Pair of nodes that encountered often in the past have high delivery predictability, and each node selects the messages that have high possibility to reach the intended destinations and transfers them to its contact node. The MaxProp routing scheme is proposed in [19], which can calculate the total cost of forwarding a message to the destination, and gives the high priority to the copy that has high delivery probability.The probabilistic routing can reduce network resource expenditure, but its performance relies on accuracy and timeliness of history information. Our proposed scheme considers both time and space factors in order to reduce the influence of the outdated history information on the routing policy.
The social network analysis is also used for DTNs routing schemes in recent years. The SimBet routing algorithm is
proposed in [20], which uses the small world dynamics to find some bridge nodes based on the centrality characteristics, and
the ‘betweenness’ and ‘similarity’ are estimated to calculate the node’s utility. On this basis, the BUBBLE scheme [21] selects
high centrality nodes and community members of destination as relays. The k-clique community detection method and
weighted network analysis are used to explore the structure of the human mobility traces. A friendship based routing in
proposed in [22], and the temporally differentiated friendships are introduced to make the forwarding decisions of messages.
The sociable routing scheme is proposed in [23], which chooses the set of best forwarders among those having high sociability
indicators, which are related to the social characteristics of networks nodes by capturing the frequency and type of their
encounters. For the social based routing schemes, how to accurately describe the social features of nodes in the dynamic DTNs
and balance the positive and negative influence of sociality are the key issues.
The controlled epidemic routing limits the number of copies injected into the networks. In [24], the spray and wait (SAW) routing algorithm is presented, in which a certain number of copies are sprayed into the networks. This algorithm waits until one of these nodes with this message meets the destination. The SAW routing can be viewed as a tradeoff between single-copy schemes and multi-copy schemes. The Spray and Focus (SAF) [25] is a modification over SAW, which is designed for a specific scenario where the node mobility is limited to a small area for most of the time. The difference between SAW and SAF lies in the second phase, while the single-copy utility based routing is performed in SAF. Based on the two algorithms mentioned above, some other spray based schemes are proposed such as time dependent message spraying [26], multi-period spraying [27], and fuzzy-spraying [28]. These schemes focused on the improvement of the spray efficiency. The spraying method used in our scheme is related to the encounter angle, which is simple but can effectively increase the message’s transmission chance. The location information is important and useful for the routing in mobile wireless networks, so it is also used to limit the amount of message copies ejected into the network.
The location-aided routing algorithms have been studied extensively following the early work in [29]. In [30], the greedy forwarding is proposed for mobile ad hoc networks, in which each node forwards the packet to the neighbor closest to the destination among all its neighbors. In [31], a stochastic location-aided routing algorithm is proposed to optimize the tradeoff between the message passing overhead and the delay. In [32], an epidemic routing protocol with H and T matrices (EPI-HT) is proposed, which tries to forward messages accurately to the next-hop according to the historical information. In [33], a spray and forward routing algorithm is proposed, which uses the Markov forecasting mechanism to predict the location of the destination node, and then forwards the message by the greedy routing. Most of these schemes need the passively updated node location database which is built by message exchange. However, since the location information exchanged in DTNs cannot be updated in time, it often leads to great difference between the real location and the stored location. In our proposed location prediction model, time is also considered as a variable to predict the location, and the current node location is related to not
only the location at the last recorded moment but also the length of the time interval. When two nodes are in contact, only the locations of the encountered nodes are exchanged, which can reduce error propagation and get more accurate prediction results.
3.
THE
PROPOSED
ALGORITHM
In this section, first both the maximum throughput estimation of node contact and the controlled spraying stage of the proposed LACS algorithm are introduced, then both the SMP based location prediction model and the single-copy routing stage of LACS are described, and the algorithm procedure is provided at the end.
3.1. The controlled spraying stage based on the contact
The development of positioning technology makes it possible to get mobile node positions easily, cheaply and precisely. The widely used positioning techniques include satellite positioning, wireless positioning and base station positioning. Especially for the satellite positioning, there are global position system (GPS), Galileo satellite navigation system, BeiDou satellite navigation system (BDS) and so on. Therefore, the assumption that the node is aware of its location is rational.
Due to the features of DTN, the location information exchange just happens when two nodes encounter. The contact time means how long the nodes are in the range of each other from being connected to being disconnected. It can reflect the maximum throughput in the contact duration and the throughput decreases with the time. The transmission fails if the message size is greater than the maximum throughput left. This leads to the energy and bandwidth waste. In the paper, a maximum throughput estimation method is proposed, which is carried out before each message is forwarded, and this can greatly
decrease the amount of unfinished message transmission. The maximum throughput in a contact is related to the node’s movement speed and direction as well as its transmission range and speed. Because most of the contacts last no more than few tens of seconds, the transmission scene can be considered as a two-dimensional model, and the movement speed and direction
are assumed constant in one contact time.
A B D B A b a
v
v
b a vb va vd Fig.1 The encounter model of nodesGiven two contacted nodes A and B as shown in Fig.1, with speed va and vb, transmission range ra and rb, transmission rate Ta and Tb respectively, the contact angle
between the two nodes can be calculated as follows:arccos a b
a b
v v v v
(1) Here, the angle
[0, ]
. Then, the contact duration time td can be calculated as follows:2(a b) cos 2(a b)( a b) d a b a b a b r r r r v v t v v v v v v
(2)From Eq.(1) and Eq.(2), the maximum throughput Cmax of this contact can be calculated as follows:
max d s
C t T (3)
Certainly, Cmax decreases gradually after the contact begins. Then, assume that the contact begins at the moment tb, and the
remaining maximum throughput
m
t
C at the moment tm can be calculated as follows:
( ( ))
m
t d m b s
C t t t T
(4) Once the maximum throughput has been obtained, the node firstly checks whether it is greater than the size of the message. If this condition is not met, the node terminates the transmission and goes to next message.
If the message can be transmitted successfully within the contact time, the next issue is how to spray the copies into the network. The traditional binary spraying is simple but not efficient. The message should be spread into the network as soon as possible to increase the delivery ratio and decrease the transmission delay. In order to increase the message’s transmission chance in the resource constrained condition, a controlled spraying method is proposed which can adjust the spraying strategy based on the encounter angle, which is superior to the binary spraying both in the efficient and speed. In our spraying method,
if the encounter angle is between[0, 2], this means that the two encounter nodes move at a similar direction, and they may appear in close area after a period of time. In this situation, the nodes exchange fewer copies of the message between each other. On the other hand, the angle between( 2, ] means that the two nodes move at quite different directions, and they may
appear in relatively further distance later. Therefore, more copies should be exchanged to spread the message more quickly and widely. This method can reduce congestion and avoid spraying too many redundant copies.
The proposed controlled spraying stage can be described as follows. When two nodes A and B encounter, they exchange the motion information and the abstract information of all messages which is called summary vector (SV). Taking node A as an example, it computes the encounter angle and estimates the maximum throughput of this contact, and then adopts the controlled spraying based on the calculation. The amount of copies to be forwarded can be obtained by the following algorithm, i.e., Algorithm 1.
Input: Message transmission queue of node A: Qm; Maximum throughput: Ct; Number of message copies: Nr; Size of message:Sm
Output: The amount of copies to be forwarded: Nf
1: for each contact of node A 2: whileQm != null 3: Update Ct;
4: for each message m in Qm 5: if Ct > Sm && Nr > 1 then 6: if 0
2 then 7: Nf = 1; 8: elseif
2
9: Nf = [Nr / 2]; 10: end if 11: else 12: Remove m from Qm; 13: end if 14: end for 15: end while 16: end for 17: returnNf;3.2. The single-copy routing stage based on location prediction
The case that a message is forwarded with only one left copy is discussed in this section. Just one copy of the message left means that some copies have been ejected into the network, thus it is necessary to find a relay node superior over the current node. In this section, a location prediction model based on the Semi-Markov Process (SMP) is introduced to support the forwarding of the last copy.
3.2.1 The SMP based location prediction model
Some location prediction models have been proposed for the nodes in wireless networks, such as the aggressive chronological projected graph [15], the LZ-based predictors [34] and the Markov-based models [35]. These models try to predict the node’s location from recent history information. They are effective only when location data are precise and timely, and some of them need to observe the node’s movement over a long period of time. Unfortunately, the long delay and frequent disruption in DTNs make it hard to exchange history location information timely and effectively. The SMP model proposed in the paper tries to overcome this difficulty by the joint use of time and location factors, in which the node’s current location is
predicted by the latest location of the node and the corresponding time, and all the required information can be obtained from the nodes in contact.
In order to establish the location prediction model, the whole communication region is divided into some grid zones as shown in Fig.2, and the grid size depends on the node transmission range. At any given moment one node is in a certain grid zone, and the central point coordinates of the zone can be used as the approximate location of the node. It is assumed that there are n different grid zones in total, and Li(xi, yi),iÎ L1 n, is the central point coordinates of theithgrid zone, which is also used as the node position in this grid zone.Because the current location of the node is related to both its previous position and the
duration time, the node’s movement can be described as a SMP. As shown in Fig.3, the location represents node’s state, and the state transition probability is related to the last state and the duration time at that state [36].
L1 L2 Ln x y
Fig.2. Divided grids in communication region Fig.3. State sequence and transition probability
It is supposed that there is a semi-Markov process Z ={ ,Z tt ³ 0} and the Kij is defined as the kernel function of this
process as follows: 1 1 P r( , | )
( )
ij n n n n ij ij K = L+ = j T + - T £ t L = i=
P H t
(5) 1 P r( | ) ij n n P L j L i += = = is the transition probability from state i to state j. The matrix P = {Pij} is used to denote the
transition probability matrix of the random process and is formulated as follows:
11 1 1 1 1 j n i ij in n nj nn P P P P P P P P P P (6)
Hij is used to denote the residence time distribution from state i to state j and it is defined as follows:
Hij = P r(Tn+1- Tn £ t L| n = i L, n+1= j)
(7) Di(t) is used to denote the duration time distribution at state i under the condition where next state is not considered. It is defined as follows: 1 ( ) P r( | ) ( ) i n n n ij j D t = T + - T £ t L = i =
å
K t (8) In fact, it is difficult to obtain Di(t) of the destination node directly. However, the residence time distribution of the current node can be used to estimate the residence time distribution of the destination node. It is described as normal distribution and the parameters are achieved from the records stored in the current node. Suppose that one node has l locationrecords and theduration time isti, iÎ L1 l, its residence time distribution can be expressed as follows:
i j n State sequence Transitions Pij Time 1
2 2 ( ) 1 ( ) ~ exp( ) 2 2 i x D t (9) Here, 1 1 l i i t l
, 2 1 1 ( ) l i i t l
.Then the transient distribution of the semi-Markov process is denoted as
f
ij( )
t
and it is defined as follows:0 ( ) P r( | ) ij t Zt j Z i f = = = 0 1 (1 - ( )) ( ) ( ) s t i ij lj il l D t d f t t dK t = = +
å ò
1 0 (1 - ( )) ( ) ( ) s t i ij il lj l D t d K t f t t dt = ¢ = +å ò
(10)Here,dijdenotes the Kronecker symbol, which is a function of two variables. Its value is 1 if the two variables are equal, and 0
otherwise. It is defined as follows:
0 1 ij i j i j d = ìïïïíï ¹ = ïïî The first part of Eq.(10) is the probability that the system does not have transition up to time t given that it was in state i at time 0. In the second part of Eq.(10), ( )
il
K¢t is the derivative at timet of ( )
il
K t and is the probability that the system remains in
state i up to timet and transits into state l just at timet . After the transition from state i to l, the system transits into state j following one of the possible trajectories from state l to state j.
By means of the quadrature method, it is easy to find a numerical solution of Eq. (10) which converges to the discrete time semi-Markov process [37]. By performing the discretization of Eq. (10) with step h, we can get the followings:
1 1 ( ) (1 ( )) ( ) (( ) ) s k h h ij i ij il lj l kh D kh v h k h t f d t f t = = = - +
å å
(11)We further get the approximateK khil¢( )in the expression ofv khil( )as follows:
( ) 1 ( ) ( ) ( ) (( 1) ) 1, il il il il il K h k v kh hK kh K kh K k h k ìï = ïï ¢ = = íï - - > ïïî % % % Where ( ) il
K kh% is the empirical distribution of ( )
il
K t , obtained from the historical data.
In fact,
f
ij(
kh
)
is the transition probability of a node from state i to state j, and the node has been in state i lasting kh time.Thefij( )k can be derived from the iterative method shown in Eq. (12).
( ) ( ) (( ) ) ( ) 1 k h h h h kh Vth k t h Dkh t f f -= -
å
= (12)Then, we can get the followings: h h h U ×F = D (13) Here, 2 3 2 1 0 0 0 1 0 0 1 0 1 h h h h h h h h h h h h h V V V U V V V é ù ê ú ê- ú ê ú ê ú - -= ê ú ê ú ê- - - ú ê ú ê ú ê ú ë û K K K K M M M M O , 0 2 3 h h h h h h h h f f f f é ù ê ú ê ú ê ú ê ú F = ê ú ê ú ê ú ê ú ê ú ê ú ë ûM , 0 2 3 h h h h h h h h D D D D D é ù ê ú ê ú ê ú ê ú = ê ú ê ú ê ú ê ú ê ú ê ú ëMû . Obviously, 0 1 h f = , and h h f can be obtained by h h h h h Vh h Dh f = ×f + . Once 0 h f , h h f ,K, h k
f are known, we can get the
followings: 1 (( 1) ) ( ) (( 1 ) ) (( 1) ) 1 k h h h h k h Vth k t h D k h t f f + + + - + = =
å
+ (14)So the transient distribution probability matrix F ={ }fij is calculated from Eq. (13) and denoted as follows:
11 1 1 1 1 j n i ij in n nj nn (15)
If the matrixis known, the location of a node can be predicted by finding the largest entry in the ith row. The
corresponding column for this entry is the most possible location that the node will be after timeDt.
3.2.2 The single-copy routing based on location prediction
If the message to be forwarded has only one copy left, the proposed scheme enters the single-copy routing stage. In this case the message should be forwarded to a relay node superior over the current node. For the contact nodes, the superior node is the one which can meet the destination node earlier.In this section the location information is used to estimate the time one node need before having contact with the destination node, and whether to forward the message or not is judged with the aid of the estimation.
Suppose that the locations of contact nodes A and B are La (xa, ya) and Lb (xb, yb), and the predicted location of destination
node D is Ld (xd, yd). The vectors between the contact nodes and destination node are calculated byA D = (xd- x ya, d- ya) uuur
,
( d b, d b)
BD = x - x y - y
uuur
arccos a ad a AD v AD v , arccos b bd b BD v BD v . (16)
Then the needed time tad and tbd for the contact nodes to encounter destination node can be estimated as follows:
2 2 ( ) ( ) cos a d a d ad a ad x x y y t v , 2 2 ( ) ( ) cos b d b d bd b bd x x y y t v (17)
Whether or not to forward the message from node A to node B is based on the comparison of tad and tbd. If tad > tbd, node A will forward this message to B, otherwise it will do nothing and process the next message.Algorithm 2 gives the detailed procedure of single-copy routing stage.
Algorithm 2 Single_copy_routing
Input: Message transmission queue of node A: Qm; Maximum
throughput Ct; Number of message copies: Nr; Size of message: Sm; Location vector: (L1, L2, …, Ln) ; Time vector (t1, t2, …, tn );
Output: The amount of copies to be forwarded: Nf
1: for each contact of node A 2: whileQm != null 3: Update Ct;
4: for each message m in Qm 5: if Ct > Sm then
6: whileNr == 1
7: Replace Ld (xd, yd) by SMP prediction model; 8: Calculate tad and tbd;
9 if tad > tbd then 10: Nf =1; 11: else tad < tbd 12: Nf =0; 13: end if 14: end while 15: else 16: Remove m from Qm; 17: end for 18: end while 19: end for 20: returnNf;
3.3. The location-aided controlled spraying routing
The features of the proposed LACS routing algorithm have been described in the above sections, and the following is the detailed description. It assumes that the nodes adopt the First Input First Output (FIFO) buffer management policy.
If two nodes are in contact at one moment in the LACS routing scheme, they firstly exchange routing related information, such as location matrix, speed, index of SV.Suppose node A is the current node and node B is the target node. In order to form the message transmission queue, node A drops the message which is outdated or has been transmitted. The size of each message in the queue is compared to the remaining throughput of the contact before forwarding. When the throughput meets the transmission requirements, the LACS enters the controlled spraying stage or the single-copy routing stage according to the amount of the message copies. The angle related adaptable spraying is introduced in the controlled spraying stage, and the SMP based location prediction is used in the single copy routing stage. For the target node B, it deals with the received message based on the message type. The transmission queue of node B will be updated after receiving acknowledgment message, and the data message will be stored and an acknowledgment message will be sent to node A. The detailed procedure of the LACS scheme is described in Algorithm3.
Algorithm 3 LACS routing algorithm
Input: Encountered nodes(A, B) ;
Output: The amount of copies to be forwarded: Nf
1: for each contact between A and B 2: Exchange routing information;
3: Establish message transmit queue of node A: Qm
4: Calculate encounter angle: arccos a b
a b
v v v v
；5: Get the maximum throughput of node A in the contact: max d a
C t T ；
6: whileQm != null
7: for each message m in Qm
8: Update the remaining maximum throughput:
( ( )) m t d m b a C t t t T ； 9: if m t C > Smthen 10: ifNr > 1 then
11: Goto Controlled_spraying_routing algorithm; 12: elseif Nr == 1
13: Goto Single_copy_ routing algorithm; 14: end if 15: else 16: Remove m from Qm; 17: end if 18: end for 19: end while 20: end for 21: return Nf
4.
SIMULATION
AND
ANALYSIS
The simulation experiments are implemented over the ONE simulator [38]. It allows users to create scenarios based upon different synthetic movement models and real-world traces and offers a framework for implementing routing and application protocols. The Epidemic, SAW and MaxProp are used as performance comparison benchmarks. The Epidemic is a fundamental routing protocol for DTNs, and the SAW is a typical spraying scheme by limiting the number of message copies forwarded, and the MaxProp is a probability routing scheme based on prioritizing messages to be transmitted and dropped. The metrics used in the performance evaluation include average message delivery ratio, end-to-end message delay and network overhead ratio. The network overhead ratio is defined as the ratio of the number of forwarded messages to the messages successfully delivered.
4.1. Simulation results in RWP movement model
In this section, the LACS, Epidemic, SAW and MaxProp are simulated with RWP movement model. The total number of nodes in the network is 200. The whole simulation time is 45000 seconds, and the initialization time is 1800 seconds. The detailed simulation parameter settings are summarized in Table I.
TABLE 1 Simulation parameters
Parameter Value
Simulation area 4, 5003, 400 m
Node bandwidth 2 Mbps
Node velocity 0.5 ~ 1.5 m/s
Node transmit range 20 m
Message size 0.5 ~ 1 MB
Message TTL 3, 600 s
Initial energy of node 10000 mA•h Energy expenditure per scan 10 mA•h Transmit energy expenditure per sec 20 mA•h
The comparison results are shown in Figs. 4-6. As shown in Fig. 4, the LACS scheme has much better performance in delivery ratio than the SAW and Epidemic, and slightly outperforms the MaxProp. When the buffer space is smaller than 20MB, the delivery ratio of SAW increases stably, but the delivery ratios of other three schemes grow rapidly, and the LACS has a similar growth trend as the MaxProp. With the increase of the buffer space, the delivery ratio of LACS will increase slightly. The reason is that the Maxprop is a kind of probabilistic routing, and it needs more message copies to improve delivery ratio. The SAW scheme restricts the number of message copies to a fixed value, thus the buffer space has less influence on it. The LACS has message forwarding process in the second stage, thus the delivery ratio increases with the buffer
storage. The LACS adopts throughput capacity estimation and location assisted methods to enhance the purpose of the message transmission, and it always achieves better performance except some extreme cases such as a very small memory.
Fig.4. Delivery ratio vs buffer size Fig. 5. Delay vs buffer size under RWP movement model under RWP movement model
As shown in Fig. 5, the LACS scheme has the lowest end-to-end message delay among the four routing algorithms. The Epidemic has the longest transmission delay and the MaxProp also has poor performance in this respect. In SAW and LACS, the delay will increase when the buffer space is less than 20 MB and then remains nearly unchanged with the increase of the buffer space. The reason is that more message copies are forwarded with the increase of buffer size in Epidemic and MaxProp, and the message copies number are limited in SAW and LACS. The location aided method reduces the delay further.
Fig.6. Overhead ratio vs buffer size under RWP movement model
As illustrated in Fig. 6, the SAW has the best performance in the message overhead ratio because it just forwards message to the destination node in the wait phase. The LACS still forwards copy to some appropriate nodes, however, the single-copy routing method in its second stage brings the improved performance compared to Epidemic and MaxProp. The Epidemic and MaxProp do not limit the number of copies, thus they have poor performance in the message overhead ratio aspect.
4.2. Simulation results in realistic trace
In this case, the Dartmouth trace is applied in the simulation [36], which was collected by several thousand wireless laptops carried by students and faculties at the Dartmouth College campus. All users are sorted in a descending order of trace length
and 200 users are selected as the nodes for simulation. The parameters of the nodes are the same as in section 4.1, and the comparison results are shown in Figs. 7-9.
Fig.7. Delivery ratio vs buffer size under Dartmouth trace Fig.8. Delay vs buffer size under Dartmouth trace As shown in Fig. 7 and Fig. 9, the delivery ratio of the four routing schemes under Dartmouth trace is slightly less than the one under RWP movement model. However, the network overhead ratio of the four routing schemes under Dartmouth trace is approximately doubled compared to RWP movement model. The reason is that the nodes in RWP movement model have more contacts than in the realistic trace, and the contacts can increase the delivery ratio while reducing the networks overhead ratio. As shown in Fig. 8, the message transmission delay of the four routing schemes under Dartmouth trace has not changed substantially in contrast to RWP movement model. This is because that the delay is mainly related to the routing algorithm, and that the contact chance does not have much effect on the transmission delay.
Fig.9. Overhead ratio vs buffer size under Dartmouth trace
5.
CONCLUSION
To improve message forwarding efficiency in DTN routing, in this paper a novel routing scheme assisted by location information is proposed with spraying stage and single-copy routing stage. It adopts throughput estimation for every contact to avoid unnecessary message forwarding. In the spraying stage, the number of copies to be forwarded is based on the encounter angle when nodes meet. In the single-copy routing stage, whether or not to forward the last copy to another node is related to the arrival time to destination node and the time is estimated with the location prediction model based on the SMP. We analyze the prediction model and develop the method to obtain the parameters of SMP model by contact information. Simulation
results show that the proposed scheme has better performance than some typical routing algorithms for DTNs under both random movement model and realistic trace. In the future, we plan to improve the location estimation precision by finding the optimal values of the parameters in the SMP model, and refine the proposed scheme to deal with other different challenging network environments, such as throwbox-equipped DTNs [39], vehicular DTNs [40] and so on.
ACKNOWLEDGMENTS
This work is supported by the National Science Foundation for Distinguished Young Scholars of China under Grant
No. 61225012 and No. 71325002; the National Natural Science Foundation of China under Grant No. 61572123.
R
EFERENCES[1] K. Fall, S. Farrell, DTN: an architectural retrospective. J. IEEE Sel. Areas Commun., 26(5) (2008) 828-836.
[2] A.G. Voyiatzis, A survey of delay-and disruption-tolerant networking applications. J. Internet Eng. 5(1) (2012) 331-344.
[3] Y. He, C. Liu, Y. Pan, J. Zhang, A localized efficient forwarding algorithm in large-scale delay tolerant networks. In: Proceeding of International Conference on Ad Hoc and Sensor Systems (MASS), Philadelphia, Pennsylvania, United States, 2014, pp. 594-599.
[4] W. Li, Y. Hu, X. Fu, S. Lu, D. Chen, Cooperative positioning and tracking in disruption tolerant networks. J. IEEE Trans. Parall. Distr., 26(2) (2015) 382-391.
[5] T. Spyropoulos, K. Psounis, C.S. Raghavendra, Efficient routing in intermittently connected mobile networks: the multiple-copy case. J. IEEE/ACM Trans. Netw., 16(1) (2008) 77-90.
[6] T. Abdelkader, K. Naik, A. Nayak, N. Goel, V. Srivastava, A performance comparison of delay-tolerant network routing protocols. J. IEEE Netw., 30(2) (2016) 46-53.
[7] Y. Wang, S. Jain, M. Martonosi, Erasure coding based routing for opportunistic networks. In: Proceeding of Special Interest Group on Data Communication (SIGCOMM), Philadelphia, PA, United States, 2005, pp.229-236
[8] T.D. Cola, M. Marchese, Adaptive erasure coding for interplanetary networks with incomplete channel side information. In: Proceeding of Global Telecommunications Conference (GLOBECOM), New Orleans, LA, United States, 2008, pp.2875-2879.
[9] T.D. Cola, M. Marchese, Joint use of custody transfer and erasure codes in DTN space networks: benefits and shortcomings. In: Proceeding of Global Telecommunications Conference (GLOBECOM), Miami, Florida, United States, 2010, pp.1-5.
[10] S. Jain, M. Demmer, R Patra, Using redundancy to cope with failures in a delay tolerant network. In: Proceeding of Special Interest Group on Data Communication (SIGCOMM), Philadelphia, PA, United States, 2005, pp.109-120.
[11] J. Widmer, J.Y.L. Boudec, Network coding for efficient communication in extreme networks. In: Proceeding of Workshop on delay tolerant networking and related networks (WDTN), Philadelphia, PA, United States, 2005, pp.430–431.
[12] Y. Lin, B. Li, B. Liang, Stochastic analysis of network coding in epidemic routing. J. IEEE Sel. Areas Commun., 26(5) (2008) 794-808.
[13] Y. Lin, B. Li, B. Liang, Efficient network coded data transmissions in disruption tolerant networks. In: Proceeding of the Conference on Computer Communications (INFOCOM), Phoenix, Arizona, United States, 2008, pp. 2180-2188.
[14] M. Chuah, P. Yang, Y. Xi, How mobility models affect the design of network coding schemes for disruption tolerant networks. In: Proceeding of the International Conference on Distributed Computing Systems Workshops (ICDCS), Montreal, Quebec, Canada, 2009, pp. 172-177.
[15] Z. Guo, B. Wang, J H. Cui, Prediction assisted single-copy routing in underwater delay tolerant networks. In: Proceeding of Global Telecommunications Conference (GLOBECOM), Miami, Florida, United States, 2010, pp.1-6.
[16] T. Spyropoulos, K. Psounis, C.S. Raghavendra, Efficient routing in intermittently connected mobile networks: the single-copy case. J.IEEE/ACM Trans. Netw., 16(1) (2008) 63-76.
[17] Q. Yuan, I. Cardei, J. Wu, Predict and relay: an efficient routing in disruption-tolerant networks. In: Proceeding of ACM international symposium on Mobile ad hoc networking and computing (MobiHoc), New Orleans, Louisiana, United States, 2009, pp.95-104.
[18] A. Lindgren, A. Doria, O. Schelén, Probabilistic routing in intermittently connected networks. Springer Berlin Heidelberg, 2004, pp.239-254.
[19] J. Burgess, B. llagher, D. Jensen, et al. MaxProp: routing for vehicle-based disruption-tolerant networks. In: Proceeding of International Conference on Computer Communications (INFOCOM), Barcelona, Spain, 2006, pp.1-11.
[20] E. Daly, M. Haahr. Social network analysis for routing in disconnected delay-tolerant MANETs. In: Proceedings of ACM international symposium on Mobile ad hoc networking and computing (Mobihoc), Montréal, Québec, Canada, 2007, pp.32-40.
[21] H. Pan, J. Crowcroft, E. Yoneki. BUBBLE Rap: social-based forwarding in delay tolerant networks. In: Proceedings of ACM international symposium on Mobile ad hoc networking and computing (Mobihoc), Hong Kong, China, 2008, pp.241-250.
[22] E. Bulut, B. K. Szymanski. Friendship based routing in delay tolerant mobile social networks. In: Proceedings of IEEE Global Telecommunications Conference (GLOBECOM), Miami, Florida, USA, 2010, pp.1-5.
[23] F. Fabbri, R. Verdone. A sociability-based routing scheme for delay-tolerant networks. EURASIP. J. Wirel. Comm., 2011(1) (2011) 1-13.
[24] T. Spyropoulos, K. Psounis, C.S. Raghavendra, Spray and wait: an efficient routing scheme for intermittently connected mobile networks. In: Proceeding of Special Interest Group on Data Communication (SIGCOMM), Philadelphia, PA, United States, 2005, pp.252-259.
[25] T. Spyropoulos, K. Psounis, C.S. Raghavendra, Spray and focus: efficient mobility-assisted routing for heterogeneous and correlated mobility. In: Proceeding of Pervasive Computing and Communications Workshops (PerCom), White Plains, New York, United States, 2007, pp.79-85.
[26] E. Bulut, Z. Wang, B K. Szymanski, Time dependent message spraying for routing in intermittently connected networks. In: Proceeding of Global Telecommunications Conference (GLOBECOM), New Orleans, LA, United States, 2008, pp.5373-5378.
[27] E. Bulut, Z. Wang, B.K. Szymanski, Cost-effective multiperiod spraying for routing in delay-tolerant networks. J. IEEE/ACM Trans. Netw., 18(5) (2010) 1530-1543.
[28] A. Mathurapoj, C. Pornavalai, G. Chakraborty, Fuzzy-spray: efficient routing in delay tolerant ad-hoc network based on fuzzy decision mechanism. In: Proceeding of IEEE International Conference on Fuzzy Systems (FUZZ), Jeju Island, Korea, 2009, pp.104 - 109.
[29] Y. Ko, N. H.Vaidya, Location-aided routing (LAR) in mobile ad hoc networks. J. Wirel. Netws., 6(6) (1998) 66-75.
[30] X. Yang, J. Yin, S. Yuan, Location-aided opportunistic routing for mobile Ad Hoc networks. In: Proceeding of International Conference on Wireless Communications, NETWORKING and Mobile Computing (WiCOM), Beijing, China, 2009, pp.1-5.
[31] Y. Zhu, J. Zhang, K. Partel, Stochastic location-aided routing for mobile ad-hoc networks. In: Proceeding of IEEE International Conference on Wireless Communications, Networking and Information Security (WCNIS),Beijing, China, 2010, pp. 364-370.
[32] Q. Liu, L. Wang, Y. Li, Y.Pan, An improved protocol based on location information in DTN. In: Proceeding of IEEE International Conference on Future Computer and Communication (FCC), Shanghai, China, 2010, pp. 520-524.
[33] F. Dang, X. Yang, K. Long, Spray and forward: efficient routing based on the Markov location prediction model for DTNs. J. Sci. China Inform. Sci., 55(2) (2012) 433-440.
[34] J. Ziv, A. Lempel, Compression of individual sequences via variable-rate coding. J. IEEE Trans. Info. Theo., 5(24) (1978) 530–536.
[35] L. Song, D. Kotz, R. Jain, X. He, Evaluating next-cell predictors with extensive Wi-Fi mobility data. J. IEEE Trans. Mob. Comput., 5(12) (2006) 1633-1649.
[36] S.Z. Yu, Hidden semi-Markov models. J. Artif. Intell., 174(2) (2010) 215-243.
[37] G. Corradi, J. Janssen, R. Manca, Numerical treatment of homogeneous semi-Markov processes in transient case–a straightforward approach. J. Methodol. Comput. Appl., 6(2) (2004) 233-246.
[38] A. Keränen, T. Kärkkäinen, J. Ott, Simulating mobility and DTNs with the ONE. J. Commun., 5(2) (2010) 92-105.
[39] D. Zeng, C. Teng, H. Yao, et al, Stochastic analysis of epidemic routing based anycast in throwbox-equipped DTNs. In: Proceeding of IEEE International Conference on Embedded Multicore/Manycore Socs (MCSoC), Aizu-Wakamatsu, Japan, 2014, pp. 77-81.
[40] J. Zhou, X. Dong, Z. Cao, et al. Secure and privacy preserving protocol for cloud-based vehicular DTNs. J. IEEE Trans. Info. Fore. Sec., 10(6) (2015) 1299-1314.
